Deployment & Ops

systemd process supervision, paper-to-live promotion checklists, infrastructure as code for trading hosts, canary releases, chaos engineering, secrets vaults.

Skills: 30
Hand off: 216
Handed off from: 197

Index

All 30 skills in this domain, alphabetically. Hover a row for its situation; open it for the full playbook.

A

audit-logging-for-configuration-changesUse when a change to a trading parameter or risk-control setting must leave evidence of who changed what, from which value to which, and why. A hash-chained record, not an authorisation gate; manual overrides belong to risk-control-bypass-audit-logging.audit-logging automated-rollback-triggers-on-anomaly-detectionUse immediately after deploying a trading algorithm, to watch post-deployment health metrics and roll back automatically on a deployment defect, with flapping and rollback-loop protection. Not for strategy underperformance.ci-cd

B

blue-green-deployment-for-live-strategy-updatesUse when a strategy that is currently routing orders and holding positions must be replaced without waiting for a maintenance window; stages the new build on a standby slot, health-checks it, syncs position and open-order state, then cuts routing authority over.blue-green

C

canary-releases-for-strategy-code-changesUse when new or materially changed strategy code will send live orders and the first ones should be small enough that a defect costs a rounding error; shadow mode first, then deliberately shrunken orders under absolute notional caps.canary capacity-planning-for-symbol-universe-growthUse before expanding a strategy from a small basket to a large universe, to forecast CPU, memory and network bandwidth from per-symbol measurements. A first-order procurement estimate, not a load test.throughput centralized-secrets-management-vault-integrationUse when a trading process should fetch exchange API keys or database credentials from HashiCorp Vault at runtime rather than a .env file; AppRole login, KV v2 reads, explicit token TTL handling and a bounded local cache.hashicorp chaos-engineering-for-trading-infrastructureUse when failover, timeout and gap-recovery paths have never actually run under fault; injects reproducible latency, message loss and simulated process death at an I/O boundary, behind an activation gate and never near live capital.chaos-engineering configuration-drift-detection-across-environmentsUse before a configuration tree takes effect somewhere that can send orders, to prove it matches the approved baseline; catches missing keys, changed values and silent type changes carried over from a debugging session.deployment-ops cost-monitoring-for-cloud-trading-infrastructureUse when cloud spend on compute, egress and market data needs a spike alarm and unit-economics view; rolling z-score baselines over telemetry you supply, reporting rather than terminating anything.cloud-cost

D

database-backup-and-point-in-time-restore-testingUse when a trading database must be provably restorable to an exact timestamp before a bad migration; replays archived WAL metadata to a target time, detects gaps and unreachable targets, and audits measured RPO and RTO.pitr-restore dependency-pinning-and-reproducible-buildsUse when auditing a requirements file before it is used with pip hash-checking mode, to verify exact == pins and valid strong hashes. It audits; resolving versions is pip-compile or uv with generate-hashes.reproducible-builds dependency-vulnerability-scanning-in-ciUse as the CI policy gate behind a real scanner such as pip-audit or OSV-Scanner: matches a dependency inventory against advisories using PEP 440 ordering and fails the build on critical or high findings.ci-cd-security deployment-freeze-windows-around-market-eventsUse in a release pipeline to block production deploys inside macro-event windows such as FOMC, CPI, payrolls and expiry, and inside timezone-aware session windows, with a named two-person override.market-events disaster-recovery-runbook-for-full-region-outageUse when a trading stack spans two cloud regions and someone must decide under time pressure whether to move it; sequences outage verification, order cancellation, database promotion, DNS switchover and reconciliation, refusing to resume trading on unverified state.region-failover

E

environment-parity-dev-staging-productionUse as the gate when a build moves between dev, staging and production, auditing declared runtime release, dependency lockfile hash, schema head and broker endpoint. Numeric research-versus-live signal parity is research-environment-vs-production-environment-parity.environment-parity

G

graceful-degradation-priority-during-partial-outageUse when a trading system is degraded but still running and must decide which work to stop doing first; a four-tier priority hierarchy that keeps risk and cancel traffic ahead of analytics and logging.load-shedding

I

immutable-infrastructure-for-trading-botsUse before a container that can send orders reaches a host, to check the running code cannot be edited in place: read-only root filesystem, pinned digest rather than a tag, no interactive shell path, and a signature attestation.immutable-infrastructure infrastructure-as-code-for-trading-hostsUse when a co-located or bare-metal trading host is provisioned as code, to validate a CPU isolation, C-state, socket buffer and PTP specification before rendering it into Terraform and Ansible. It audits a spec, not a live machine.terraform

L

load-testing-before-scaling-to-new-instrument-universeUse as the gate before expanding an instrument universe, projecting tick throughput, order-book memory, bandwidth and database write IOPS from measured per-symbol figures to decide whether a replay test is warranted.universe-scaling log-aggregation-and-centralized-observabilityUse when distributed trading services ship logs off-host to Loki, an OpenTelemetry collector or ELK; redacts credential-bearing keys before they leave, and emits structured JSON with trace correlation.observability

M

multi-region-failover-for-broker-connectivityUse when a bot reaches its broker over more than one path and something must decide unattended which path carries orders; refuses to fail over onto a path nobody has probed. Failing over to a different broker account is broker-failover-secondary-account-routing.failover

N

network-segmentation-for-trading-infrastructureUse when auditing whether an attacker landing in the least-trusted subnet can reach an order gateway or a signing host; checks declared subnets and firewall rules for direct and multi-hop paths.network-segmentation

O

on-call-rotation-and-escalation-for-trading-systemsUse when designing on-call rotations and escalation for live trading, where an unacknowledged kill-switch alert means a position running unsupervised; shift-aware resolution, severity ladders and acknowledgement SLA auditing.sre

P

paper-to-live-promotion-checklistUse as the final gate before a strategy routes real orders with real capital, scoring six conjunctive criteria including paper duration, trade count and slippage alignment. Every input is an observation you supply, not a verification.paper-trading post-mortem-culture-and-blameless-review-processUse when an incident narrative has been drafted and is about to become a retained record, to screen its free-text sections for personal attribution and blame language before it is filed.sre

R

runbook-automation-for-common-incident-typesUse when monitoring already classifies a trading incident and the first minute of response should be a pre-approved sequence rather than improvisation. It executes a runbook; it does not decide whether the alert is real.incident-response

S

secrets-rotation-without-bot-downtimeUse when a long-running trading process must swap an API key or database credential without restarting: pre-swap validation, an atomic hot-swap under a lock, and a lease-gated overlap for in-flight requests. Not for OAuth refresh tokens.zero-downtime structured-logging-for-post-incident-forensicsUse when designing the log records a trading system writes so an incident can be reconstructed by query rather than by reading prose: one JSON object per event with a correlation id linking an order end to end.logging systemd-supervision-for-trading-botsUse when a bot that can send orders runs as a systemd unit and supervision must be correct rather than merely present; audits directives placed in the wrong section, restart limits that never engage, and missing hardening.systemd

Z

zero-downtime-database-schema-migrationsUse when altering the schema of a database a 24/7 trading system reads or writes, without a trading-hours outage; a five-phase expand-contract sequence with lock-bounded non-blocking DDL for Postgres and MySQL.database-migration

Search the whole catalog